Optimal Allocation Model for Social Capital Resources in Community Health Management Based on Multi-Agent Deep Reinforcement Learning

Abstract

In community health management resource allocation, the difficulty in quantifying social capital and the inefficiency of multi-agent collaboration lead to imbalanced allocation. This problem, mirrored in the textile industry’s complex supply chain where poor coordination and unquantifiable supplier relationships cause imbalanced raw material and production allocation, is addressed by this paper. We propose a model integrating dynamic social network embedding with hierarchical multi-agent deep reinforcement learning (MADRL). A graph neural network extracts node embeddings from time-series interaction data to achieve a quantitative representation of social capital. A hierarchical multi-agent architecture is constructed using heterogeneous agents responsible for demand response and resource scheduling, respectively. A joint reward function integrates health improvement, resource utilization, and network fairness, while a counterfactual baseline mechanism allocates local credit to improve policy learning accuracy. Furthermore, a parameter-sharing MAPPO (Multi-Agent Proximal Policy Optimization) algorithm with entropy regularization is employed under a centralized training and distributed execution framework to achieve stable collaborative decisionmaking. Beyond community governance, the proposed dynamic graph representation and multi-agent optimization strategy provides a computational framework for distributed information coordination and adaptive resource scheduling in intelligent electromagnetic sensing and communication environments, where reliable network interaction and efficient information propagation are critical. Experimental results demonstrate that the proposed model achieves an average Gini coefficient of 0.18–0.29 for normalized healthcare demand satisfaction and a social capital activation rate of 79.3%±3.6%, significantly improving allocation fairness and utilization efficiency. By quantifying social capital and enhancing multi-agent collaboration, this study provides an effective framework for alleviating distribution imbalance while offering methodological insights for intelligent resource management in complex networked engineering systems.